class InferenceEngine::CNNNetwork¶

Overview¶

This class contains all the information about the Neural Network and the related binary information. More…

#include <ie_cnn_network.h>

class CNNNetwork
{
public:
    // construction

    CNNNetwork();
    CNNNetwork(std::shared_ptr<ICNNNetwork> network);

    CNNNetwork(
        const std::shared_ptr<ngraph::Function>& network,
        const std::vector<std::shared_ptr<IExtension>>& exts = {}
        );

    // methods

    OutputsDataMap getOutputsInfo() const;
    InputsDataMap getInputsInfo() const;
    size_t layerCount() const;
    const std::string& getName() const;
    void setBatchSize(const size_t size);
    size_t getBatchSize() const;
    operator ICNNNetwork::Ptr ();
    operator ICNNNetwork & ();
    operator const ICNNNetwork & () const;
    std::shared_ptr<ngraph::Function> getFunction();
    std::shared_ptr<const ngraph::Function> getFunction() const;
    void addOutput(const std::string& layerName, size_t outputIndex = 0);
    ICNNNetwork::InputShapes getInputShapes() const;
    void reshape(const ICNNNetwork::InputShapes& inputShapes);
    void serialize(const std::string& xmlPath, const std::string& binPath = {}) const;
    void serialize(std::ostream& xmlBuf, std::ostream& binBuf) const;
    void serialize(std::ostream& xmlBuf, Blob::Ptr& binBlob) const;
    std::string getOVNameForTensor(const std::string& orig_name) const;
};

Detailed Documentation¶

This class contains all the information about the Neural Network and the related binary information.

Construction¶

CNNNetwork()

A default constructor.

CNNNetwork(std::shared_ptr<ICNNNetwork> network)

Allows helper class to manage lifetime of network object.

Deprecated Don’t use this constructor. It will be removed soon

Parameters:

network

Pointer to the network object

CNNNetwork(
    const std::shared_ptr<ngraph::Function>& network,
    const std::vector<std::shared_ptr<IExtension>>& exts = {}
    )

A constructor from ngraph::Function object This constructor wraps existing ngraph::Function If you want to avoid modification of original Function, please create a copy.

Parameters:

network	Pointer to the ngraph::Function object
exts	Vector of pointers to IE extension objects

Methods¶

OutputsDataMap getOutputsInfo() const

Gets the network output Data node information. The received info is stored in the given Data node.

For single and multiple outputs networks.

This method need to be called to find out OpenVINO output names for using them later when calling InferenceEngine::InferRequest::GetBlob or InferenceEngine::InferRequest::SetBlob

If you want to use framework names, you can use InferenceEngine::CNNNetwork::getOVNameForTensor method to map framework names to OpenVINO names

Returns:

the InferenceEngine::OutputsDataMap object

InputsDataMap getInputsInfo() const

Gets the network input Data node information. The received info is stored in the given InputsDataMap object.

For single and multiple inputs networks. This method need to be called to find out OpenVINO input names for using them later when calling InferenceEngine::InferRequest::SetBlob

If you want to use framework names, you can use InferenceEngine::ICNNNetwork::getOVNameForTensor method to map framework names to OpenVINO names

Returns:

The InferenceEngine::InputsDataMap object.

size_t layerCount() const

Returns the number of layers in the network as an integer value.

Returns:

The number of layers as an integer value

const std::string& getName() const

Returns the network name.

Returns:

Network name

void setBatchSize(const size_t size)

Changes the inference batch size.

There are several limitations and it’s not recommended to use it. Set batch to the input shape and call InferenceEngine::CNNNetwork::reshape.

Current implementation of the function sets batch size to the first dimension of all layers in the networks. Before calling it make sure that all your layers have batch in the first dimension, otherwise the method works incorrectly. This limitation is resolved via shape inference feature by using InferenceEngine::ICNNNetwork::reshape method. To read more refer to the Shape Inference section in documentation

Parameters:

size	Size of batch to set

size_t getBatchSize() const

Gets the inference batch size.

Returns:

The size of batch as a size_t value

operator ICNNNetwork::Ptr ()

An overloaded operator cast to get pointer on current network.

Deprecated InferenceEngine::ICNNNetwork interface is deprecated

Returns:

A shared pointer of the current network

operator ICNNNetwork & ()

An overloaded operator & to get current network.

Deprecated InferenceEngine::ICNNNetwork interface is deprecated

Returns:

An instance of the current network

operator const ICNNNetwork & () const

An overloaded operator & to get current network.

Deprecated InferenceEngine::ICNNNetwork interface is deprecated

Returns:

A const reference of the current network

std::shared_ptr<ngraph::Function> getFunction()

Returns constant nGraph function.

Returns:

constant nGraph function

std::shared_ptr<const ngraph::Function> getFunction() const

Returns constant nGraph function.

Returns:

constant nGraph function

void addOutput(const std::string& layerName, size_t outputIndex = 0)

Adds output to the layer.

Parameters:

layerName	Name of the layer
outputIndex	Index of the output

ICNNNetwork::InputShapes getInputShapes() const

Helper method to get collect all input shapes with names of corresponding Data objects.

Returns:

Map of pairs: input name and its dimension.

void reshape(const ICNNNetwork::InputShapes& inputShapes)

Run shape inference with new input shapes for the network.

Parameters:

inputShapes

A map of pairs: name of corresponding data and its dimension.

void serialize(const std::string& xmlPath, const std::string& binPath = {}) const

Serialize network to IR and weights files.

Parameters:

xmlPath	Path to output IR file.
binPath	Path to output weights file. The parameter is skipped in case of executable graph info serialization.

void serialize(std::ostream& xmlBuf, std::ostream& binBuf) const

Serialize network to IR and weights streams.

Parameters:

xmlBuf	output IR stream.
binBuf	output weights stream.

void serialize(std::ostream& xmlBuf, Blob::Ptr& binBlob) const

Serialize network to IR stream and weights Blob::Ptr.

Parameters:

xmlBuf	output IR stream.
binBlob	output weights Blob::Ptr.

std::string getOVNameForTensor(const std::string& orig_name) const

Method maps framework tensor name to OpenVINO name.

Parameters:

orig_name

Framework tensor name

Returns:

OpenVINO name